In this study, we have presented an ECV-guided LGE analysis method that uses the ECV map as a guide to determine the optimal LGE n-SD threshold in individual cases. This method was developed in a cohort of women with or at-risk of HIV which presented an LGE pattern of diffuse and patchy non-ischemic cardiomyopathy. The ECV cutoff of 31.5% successfully differentiated scar from non-scar, achieving high sensitivity, specificity, NPV, and PPV. LGE n-SD threshold was optimized using the ECV map as a reference, which also contributed to the high reproducibility of this method. The selected n-SD threshold ranged from 3SD to 18SD. The current study also suggested that the manual LGE assessment on non-ischemic diffuse and patchy fibrosis may be overestimating the scar amount. Overall, the ECV-guided LGE analysis was a more robust LGE quantification than the conventional quantification method for non-ischemic LGE cases. The robustness of ECV-guided LGE analysis was also confirmed in two validation cohorts of non-ischemic and ischemic cases.
The development of ECV criteria involved three steps of considerations. First, we selected ECV, not nT1, to guide the LGE analysis. This was based on the better performance of ECV to differentiate scar/non-scar than nT1 (Fig. 4). In addition, the correlation between ECV and SI was stronger than that of between nT1 and SI (ECV and SI: rs = 0.35, P < 0.01; nT1 and SI: rs = 0.27, P = 0.02), which was in line with the literature that reported a significant linear correlation between ECV and histological collagen volume fraction (CVF), but no significant correlation between nT1 and CVF [32, 43–46]. The greater robustness of ECV relative to nT1 values at different imaging parameters and scanner field strengths [33, 34] further supported our decision. Second, there was a discussion on the ECV cutoff values which corresponded to the LGE. Such ECV cutoff values specific for the HIV patients have not been investigated so far and therefore, the ECV criteria was developed from our 80 training cases. Indeed, our ECV cutoff of 31.5% was consistent with other publications. In a cohort with myocardial infarction or hypertrophic cardiomyopathy (HCM) patients, the ECV cutoff value for LGE was 32% . In a diastolic cardiomyopathy cohort, CVF cutoff of 12% (which calculates to the ECV value of 30.5%) corresponded to the LGE  while in HCM, the CVF cutoff was 15% . ECV values of remote myocardium and LGE scar area were also referenced from HCM (28 ± 4% vs. 30 ± 5 %, P < 0.001), non-ischemic cardiomyopathy (26 ± 3% vs 37 ± 6%, P < 0.001), and myocardial infarction (27 ± 3% vs 51 ± 8%, P < 0.001). Thirdly, there was a potential trade-off of false-positive or false-negative with regard to the single cutoff strategy. Such misclassification was observed in 9 cases (11.3%), typically when ECV values were close to the ECV cutoff value of 31.5% (averaged ECV was 31.6 ± 1.9%). A lower ECV cutoff to achieve a higher sensitivity or a higher ECV cutoff to achieve a higher specificity may be considered, although in such situations, the counterpart of specificity or sensitivity will be compromised.
Our study presented multiple advantages of ECV-guided LGE analysis for the non-ischemic LGE quantification. First, the personalized optimization of the LGE n-SD cut-off enables its application to different pathophysiologies and to assess disease progression. Considering the broad range of n-SD threshold applied, a fixed semi-quantitative threshold was not the optimal choice for our cohort. This finding is in line with the publication that the individual optimization of the LGE cut-off was more effective than a fixed cut-off of 2-SD or 6-SD in a cohort of HCM . Second, given the ECV map as a guide, the observers could rationally determine the absence of a scar (Fig. 3, Case 2). In our study, this contributed to the difference in scar detection rate or scar amount between the proposed method and the manual analysis. Utilization of the ECV map as a guide especially helps the analysts when quantifying the LGE scar with a relatively unknown distribution [6, 7, 12–16]. Third, the excellent reproducibility of ECV-guided LGE analysis is an advantage in the systematic detection of small changes in scar size for the monitoring and management of non-ischemic patients, as well as to determine the prognostic risk of patients more accurately. Many non-ischemic disease groups present relatively small LGE scar amounts as compared to ischemic cardiomyopathy and HCM [1–4]. The clinical impact of its per-unit change of LGE may likely be different among etiologies. In this regard, ECV-guided LGE analysis is sensitive to a small change in LGE scar size so that the corresponding change in myocardial disease may be detected more sensitively. Additionally, reproducibility is a key determinant of required sample sizes for clinical trials. ECV-guided LGE analysis potentially allows a substantial reduction in the sample size, which is a great benefit for a clinical study [20, 52, 53].
The validation studies on the non-ischemic and ischemic cohorts proved the robustness of the ECV-guided LGE analysis. The non-ischemic validation cohort presented similar trends as the training cohort in the inter-method analysis of the scar (%). This again suggested that the manual LGE assessment on non-ischemic cases may be overestimating the scar amount. In the ischemic cohort, ECV-guided LGE analysis achieved high reproducibility. Although in this cohort, the conventional method of FWHM with manual correction was already presenting excellent performance and therefore, there was only a small room for the proposed method to improve reproducibility. This was derived from the high SI and the well-known distribution of the ischemic scar. The ECV-guided LGE analysis was feasible in the ischemic cases but did not necessarily surpass the conventional method.
Several study limitations deserve discussion. First, ECV-guided LGE analysis cannot be performed when either the LGE or T1 map image is missing or in cases of poor image quality, which occurred in 19 cases (19%) and were excluded from this study. In such cases, manual scar analysis was performed. Secondly, the n-SD threshold selection was performed with only one slice in the mid-LV due to the current technical limitation of T1 mapping. 3D T1 mapping methods [54, 55] with a matched slice position may give a more defined n-SD threshold selection at each slice level, which may then be applied to higher resolution LGE images. Thirdly, the performance of ECV criteria was not compared with pathology, since myocardial biopsy was not available in the study protocol as well as obtaining the myocardium from the same location as suggested in the MRI images was not practical. Finally, our ECV-guided LGE method was developed in a unique cohort of women with or at risk for HIV infection and validated in a non-ischemic and an ischemic cohort but with a small number of cases. Further validation in a larger number of participants with multiple etiologies may help validate the method further.
In conclusion, ECV-guided LGE analysis is a robust and comprehensive method of scar burden and distribution assessment in participants with diffuse and patchy fibrosis, achieving both higher intra- and inter-observer reproducibility as compared to manual analysis.